Resolve DataFrame type through intermediate generic classes

[yusuke14tanaka]

Problem

get_dataframe_type_from_task infers the DataFrame backend (pandas / polars / polars-lazy) from the TaskOnKart[T] type parameter to choose the matching file processor. It inspects the type argument of a TaskOnKart[...] base directly, so when the concrete type is bound through an intermediate generic class, the TaskOnKart base still carries an unbound TypeVar and detection silently falls back to pandas.

This shows up with a common pattern where a library defines a generic base task and the concrete type is bound on a subclass:

T = TypeVar('T')

class MyBase(TaskOnKart[T], Generic[T]):
    ...

class MyTask(MyBase[pl.DataFrame]):  # detected as pandas before this change
    ...

A polars task detected as pandas means its cache is read/written with the pandas processor.

What this changes

Walk the MRO to collect TypeVar → type-argument bindings, then resolve the TaskOnKart type argument through that map (following TypeVar → TypeVar chains across multiple generic levels). Direct and nested cases are unaffected, since a non-TypeVar argument resolves to itself.

Added tests cover the intermediate-generic case for both pl.DataFrame and pl.LazyFrame.

[kitagry]

Thank you!

[hirosassa]

LGTM